Feature extraction in satellite imagery is fundamental to the goal of gathering timely, large-area geospatial information relevant to GIScience research and beyond. There are two approaches in remote sensing to feature extraction. One approach involves identifying phenomena in imagery to be reduced into map form (typically features such as categories or land surface elements). A second approach is to enhance and extract specific bands of imagery and transform them in order to provide a reduced set of inputs or predictors to a model (e.g., a vegetation index). This section focuses only on the former. Extraction of features is performed using a conceptualization of the study site known as a scene model, and the combination of ground reference information and appropriately chosen satellite data. Features can be represented in maps as discrete pixels, polygons or fuzzy membership surfaces, and machine learning algorithms have emerged as the most reliable and effective approaches to feature extraction in the last decade. There are five key steps to performing effective feature extraction: (1) developing a scene model to determine the appropriate scales of information required for a project; (2) ground reference data collection to support the calibration and validation process; (3) selection of appropriate satellite image data, and this can include ancillary data such as digital elevation models; (4) application of a feature extraction algorithm that can best distinguish the feature(s) of interest from background features and produce a map product that is logically consistent; and (5) assessment of map accuracy using validation data to determine the quality of the product for various uses.